Milan Klöwer

> ```python > >>> bin(0xffff_ffff_ffff_ffff '0b1111111111111111111111111111111111111111111111111111111111111111000000000000000000000000000000000000000000' > >>> (0xffff_ffff_ffff_ffff 106 > ``` It seems that for

Once this works, could you benchmark your binary round function? I would be interested how quickly python can process large arrays.

Thanks Zachary, I believe there is this optimization potential: Binary rounding should not allocate more memory than the output array. On my macbook air ```julia julia> using BenchmarkTools, Elefridge julia>...

Question: Your binary round function does not get compiled, right? Because that might explain the remaining x5 speed difference...

Yeah, you have to decide whether it's worth it. But as you can see, a binary rounding function does not have to be slower than squaring every element for example

I would need to know the details, but technically bitwise reproducibility is *easier* with Float32 if you compute in Float64 because you throw away 29 bits that could be different....

Thanks for the ping. I want to tackle this week after I submitted the JOSS paper!

No it's absolutely possible to use information about the linear packing that was previously applied to set some bound on the precision instead of assuming that a float32 has full...

[SpeedyWeather.jl](https://github.com/SpeedyWeather/SpeedyWeather.jl) has this now implemented as ```julia Weather is speedy: 100%|███████████████████████████████| Time: 0:00:29 (80.91 years/day) ``` Meaning the model ran at a speed such that in 1 day it would...

I think that was experimental and I seem to have removed it. However, you could add it in again, define similar to https://github.com/milankl/ShallowWaters.jl/blob/0169c122f52d764fa26a2bb31c2733a610fdb719/src/forcing.jl#L141-L151 and add it as an option here...